Percy Liang

How to govern open(-weight) foundation models has been a hot topic over the last year. This @ScienceMagazine article captures our thinking on benefits (let's not forget about this!) and risks of release model weights. We promoted the "marginal risk" framing that is now fairly

A new leaderboard for VLMs built using the HELM framework, featuring 21 datasets (including MMMU, RealWorldQA, VQAv2, GQA, Bingo, etc.), evaluating your favorite VLMs using standard prompts. All predictions available on our website for full transparency!

RT @nathanbenaich: this coming thurs, we're launching the annual @stateofaireport if you're in sf that evening, come join the launch meet…

RT @AlpayAriyak: I’m hiring for my Post-Training team at TogetherAI If you’re obsessed with data quality, generation, creating the strong…